[Dynamic Shapes] fixed dynamic shape inference #128807
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/128807
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (3 Unrelated Failures)As of commit 65099e1 with merge base 8a2fed7 ( BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
UNSTABLE - The following job failed but was likely due to flakiness present on trunk and has been marked as unstable:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
zero000064
changed the title
mikaylagawarecki
added
the
triaged
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Merge failedReason: 1 jobs have failed, first few of them are: inductor / linux-jammy-cpu-py3.8-gcc11-inductor / test (dynamic_cpu_inductor_torchbench, 1, 2, linux.12xlarge) Details for Dev Infra teamRaised by workflow job |
|
Here's the information related to the failure, inductor / linux-jammy-cpu-py3.8-gcc11-inductor / test (dynamic_cpu_inductor_torchbench, 1, 2, linux.12xlarge). This error comes from DALLE2_pytorch unable to be loaded on CPU, doesn't look like related to my code change.
|
|
DALLE2_pytorch doesn't look related to the code change, its error is NotImplementedError: DALL-E 2 Not Supported on CPU.
|
|
Just checked export/test_serdes.py::SerDesExportTestExport::test_derived_dim_1_2_serde, the following codes in export/test_export.py give some flexibility on input shape. But my code change ruins this flexibility, need to think about how to keep this flexibility. Luckily, it's Juneteenth tmr, I have some free time to investigate. |
|
After investigation, the change doesn't deter exporting a graph module in test_derived_dim_1_2 in test_export. But it only allows the usage of a tensor with dim 2 and blocks the usage of tensor with any other dim figure. This scenario is not the intention because the change intends to block graph compliling instead of its usage. |
|
Before this code change, the graph's first placeholder node represents a fake tensor, |
|
_update_var_to_range() is also called in constrain_symbol_range() when deserializing the the graph in _export, that's why replacement's effect propagates to the ExportedProgram, the potential solution might be avoiding code change in _update_var_to_range. Instead, move the code change to ShapeEnv._refine_ranges(). |
|
Moved the replacement logic to ShapeEnv._refine_ranges. Tested these test cases including reported failure ones. TORCH_LOGS="+dynamic,+dynamo" python test/dynamo/test_misc.py -k test_raise_guard_indirect_full_constraint TORCH_LOGS="+dynamic,+dynamo" python test/dynamo/test_misc.py -k test_guard_failure_fn_shape_control TORCH_LOGS="+dynamic,+dynamo" python test/dynamo/test_subclasses.py -k test_wrapper_subclass_guards_on_inner_tensor TORCH_LOGS="+dynamic,+dynamo" python test/export/test_serdes.py -k test_derived_dim_1_2_serdes |
|
@ezyang Could you review the change again? I changed the code in symbolic_shapes.py to fix test failures. |
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Made dynamic dimension indirectly bound to an integer constrained.
After each ShapeEnv._refine_ranges, check if the new ValueRange is singleton, if it is, replace the symbol.
Fixes #122307
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @ezyang